Available online 29 November 2025

Deep-learning reconstruction algorithms and photon-counting CT help reduce image noise and improve the detectability of simulated contrast-enhanced lesions on low-keV virtual monoenergetic images.

At 40 or 50 keV, the best combined results (based on both objective and subjective assessments) are obtained with the dual-layer CT and the photon-counting CT scanners.

The accuracy of iodine concentrations is in the same order of magnitude for all dual-energy CT scanners.

The purpose of this study was to compare the spectral performance of three dual-energy CT (DECT) scanners and one photon-counting CT (PCCT) scanner on virtual monoenergetic images (VMIs) at low-energy levels and on iodine maps.

A spectral phantom was scanned using one PCCT scanner and three different DECT scanners that included a rapid kV-switching CT (R-KVSCT), an ultrafast kV-switching (U-KVSCT) and a dual-layer CT (DLCT) scanner. Acquisitions were obtained with each CT system using classical abdominal and pelvic examination parameters, as well as a volume CT dose index at 11 mGy. VMI at 40/50/60/70 keV and iodine maps were reconstructed for each scanner. Noise power spectrum (NPS) and task-based transfer function (TTF) were evaluated. Detectability indexes (d') were computed to model the detection task of two contrast-enhanced lesions.

Noise magnitude decreased from 40 to 70 keV for all DECT scanners and this decrease was greater for R-KVSCT (-80.0 ± 0.1 [standard deviation (SD)] %) and less pronounced for DLCT (-14.4 ± 0.8 [SD] %) scanners. The average NPS spatial frequency (fav) values decreased from 40 to 70 keV (0.26 to 0.17 mm-1) for R-KVSCT, increased for DLCT (0.18 to 0.25 mm-1) but were similar for U-KVSCT (0.19 ± 0.002 [SD] mm-1) and PCCT (0.21 ± [SD] 0.008 mm-1) scanner. For R-KVSCT and PCCT scanners, TTF at 50 % (f50) values increased from 40 to 70 keV for both inserts. For U-KVSCT and DLCT scanners, similar f50 values were found according to energy level for both inserts. For both contrast-enhanced lesions, d' values decreased from 40 to 70 keV for PCCT, DLCT and U-KVSCT scanners. For R-KVSCT scanner, d' values peaked at 60 keV. At 40 and 50 keV, the greatest d’ values were found with DLCT and PCCT scanners.

At 40 or 50 keV, the best combined results (objective and subjective assessments) are obtained with DLCT and PCCT scanners.

Computed tomography

Dual-energy CT

Image quality

Photon-counting CT

Task-based image quality assessment

Deep learning image reconstruction

Deep learning spectral reconstruction

Energy-integrating detectors

Iterative reconstruction

Photon-counting detectors

Root-mean square deviation

Task-based transfer function

Spectral artificial intelligence reconstruction

© 2025 The Author(s). Published by Elsevier Masson SAS on behalf of Société française de radiologie.